European Economic Review 48 (2004) 851 – 881
www.elsevier.com/locate/econbase
Public expenditure, international specialisation
and agglomeration
Marius Br%ulharta;∗ , Federico Trionfettib; c
a D
epartement
d’econometrie et e conomie politique, Ecole des HEC, University of Lausanne,
CH-1015 Lausanne, Switzerland
b CEPN-CNRS, University of Paris 13, F-93430 Paris, France
c CEPII, F-75015 Paris, France
Received 9 May 2003; accepted 4 September 2003
Abstract
It is widely recognised that public-sector purchasers tend to favour domestic suppliers. We
study the consequences of such home-biased public procurement on international specialisation.
Using a general-equilibrium model with a monopolistically competitive sector, we 4nd that a
country will specialise in that sector if it has relatively large home-biased procurement (the “pull”
e7ect). Furthermore, home-biased procurement can counter agglomeration forces in that sector
and thereby attenuate the overall degree of international specialisation (the “spread” e7ect). Our
empirical analysis, conducted on input–output data for the European Union, yields supporting
evidence for the pull e7ect and some support for the spread e7ect.
c 2003 Elsevier B.V. All rights reserved.
JEL classi-cation: H5; F1; R3; R15
Keywords: Public expenditure; International specialisation; Economic geography; European Union; Input–
output analysis
1. Introduction
This study investigates the consequences of home-biased government procurement
on the patterns and the intensity of international specialisation. By “home bias” we
refer to governments’ preference for domestic over foreign suppliers irrespective of
cost and quality considerations. Discrimination by public purchasers in favour of local suppliers is a pervasive phenomenon, the motivations for which have been studied
∗
Corresponding author. Tel.: +41-21-692-3471; fax: +41-21-692-3305.
E-mail addresses: [email protected] (M. Br%ulhart), [email protected]
(F. Trionfetti).
c 2003 Elsevier B.V. All rights reserved.
0014-2921/$ - see front matter doi:10.1016/j.euroecorev.2003.09.002
852
M. Br3ulhart, F. Trionfetti / European Economic Review 48 (2004) 851 – 881
extensively. In this paper we investigate the consequences of home-biased procurement
on the location of manufacturing. First, we study whether and how public expenditure
a7ects the location of industries across countries. For this purpose we extend the model
of Helpman and Krugman (1985, Part III) to include home-biased public procurement.
Our analysis yields the prediction that, ceteris paribus, a country will tend to specialise
in the good for which it has relatively large home-biased procurement. We call this
result the “pull e7ect”. Our empirical investigation based on input–output data for EU
countries in 1970 –1985 supports this proposition: we 4nd robust evidence of a “pull
e7ect” of public expenditure on the location of manufacturing industries.
Our second focus is on the intensity of industrial concentration, i.e. on the “how
much” rather than the “where” of industrial agglomeration. To study this question we
extend the “new economic geography” model of Krugman and Venables (1995) to
include home-biased procurement. We 4nd that home-biased public expenditure, by
acting as a dispersion force, reduces the likelihood that agglomeration forces prevail;
and in the case that they prevail, public expenditure reduces the intensity of industrial
agglomeration. We call this result the “spread e7ect”. An analysis of this link in our
EU input–output dataset con4rms the presence of such a “spread e7ect” of public
expenditure on the location of manufacturing industries.
Previous work on public procurement has followed two principal paths: one was
to investigate the rationale for home bias, while the other was to study the consequences of home bias on international specialisation. The 4rst body of research, reviewed in Mattoo (1996), took a partial equilibrium approach and explained the home
bias through the political interplay between the tendering entity and domestic and foreign bidders. McAfee and McMillan (1989), for instance, focused on the design of
the optimal procurement mechanism in a setting where each bidder is better informed
about his own cost than either the government or other bidders. They 4nd that, if the
foreign 4rm’s costs are lower in expectation than those of the home 4rm, the government can minimise its expected procurement costs by granting a preference margin
to the home 4rm. The reason is that a discriminatory policy induces foreign 4rms to
reveal their private information about their costs and to prevent the most eHcient 4rms
from bidding too high. Interestingly, in their model, it is the eHciency of government
procurement that motivates the home bias. Conversely, in Branco (1994) and Vagstad
(1995), the home bias in public procurement stems from the assumption that pro4ts
of domestic 4rms enter the objective function of government while those of foreign
4rms do not, and government discrimination against foreign bidders causes pro4ts to
shift from foreign to domestic 4rms. The model of La7ont and Tirole (1991) features
bribes (“asymmetric collusion”) as the channel through which domestic 4rms’ pro4ts
enter the government’s objective function. In Weichenrieder (2001), the home bias is
motivated by the government’s desire to attract capital-intensive production and thereby
to increase local taxable capital. Naegelen and Mougeot (1998) have considered the
eHciency argument and the pro4t-shifting argument in the same model and in a variety
of informational settings. The optimal policy derived from this combined model results,
once again, in a discriminatory procurement mechanism.
We depart from this line of research, which has focused on the causes of the home
bias, by taking the home bias as exogenous and focusing on the consequences of this
M. Br3ulhart, F. Trionfetti / European Economic Review 48 (2004) 851 – 881
853
practice on international specialisation in a general-equilibrium setting. In this respect,
our study is closely related to a second line of inquiry on home-biased procurement,
which, although initiated over 30 years ago, has remained relatively underexplored. This
research programme links government procurement to international specialisation and
is originally due to Baldwin (1970, 1984). His work shows that, in a Heckscher–Ohlin
model, home bias in public expenditure is irrelevant for international specialisation.
The reason is that home-biased government procurement crowds out private demand
for domestic goods. Thus, the reduction in imports from the government is compensated by a corresponding increase in imports of the private economy. This somewhat
counterintuitive result has been investigated further by Miyagiwa (1991), who demonstrated that Baldwin’s “neutrality proposition” extends to a model of oligopoly with a
homogeneous good.
Baldwin’s analysis also established some cases in which neutrality does not hold.
One case is where the government’s domestic procurement requirement exceeds the
undistorted free-trade level of domestic production. In this situation, government expenditure of course a7ects international specialisation. This con4guration corresponds
to the corner solution of the Heckscher–Ohlin model where one country is completely
specialised and factor prices are no longer equalised. A variant of this con4guration appears in the three-sector general-equilibrium model of Weichenrieder (2001),
where governments are the sole buyers of an internationally traded homogenous good
(“tanks”). In a country for which the undistorted free-trade equilibrium would imply
zero production of tanks, domestic procurement of tanks obviously induces domestic
production of tanks. Furthermore, Weichenrieder (2001) has shown that even for a
country with positive tank production in the undistorted equilibrium and with domestic
procurement that is smaller than free-trade domestic output, home-biased tank procurement can increase domestic production, because domestic procurement replaces exports
less than one-for-one. Miyagiwa (1991) found that, in a partial-equilibrium oligopoly
model with di7erentiated goods, home-biased procurement can a7ect domestic output
as well. The reason is that when goods are di7erentiated, foreign and domestic goods
are not perfect substitutes, and home-biased government demand does not therefore
crowd out private demand for domestic goods entirely. Finally, Martin and Rogers
(1995) showed that public expenditure on trade-facilitating infrastructure could substantially a7ect international specialisation in a general-equilibrium model with trade
costs and monopolistic competition. This e7ect is the result of what the government
purchases (i.e. trade-facilitating infrastructure) rather than how the government allocates
its purchases (i.e. with or without home bias).
To our knowledge, ours is the 4rst study of the impact of home-biased expenditure
on international specialisation in a general-equilibrium model that features di7erentiated
goods and imperfect competition, and the 4rst to examine the e7ect of home-biased
expenditure on industrial agglomeration. Speci4cally, we assume 4rm-level increasing
returns and monopolistic competition in the di7erentiated-goods sector. We use the
static version of this model to study the e7ect of home-biased expenditure on the
specialisation of individual countries. A dynamic “new economic geography” version
of the model with agglomeration and dispersion forces is then employed to study
the consequences of the home bias on the intensity of industrial agglomeration across
854
M. Br3ulhart, F. Trionfetti / European Economic Review 48 (2004) 851 – 881
countries. The econometric tests that complement our theoretical analysis constitute, to
our knowledge, the 4rst empirical investigation of the sectoral output e7ects of public
expenditure in an international setting.
The analysis in this paper is of a positive nature, but it may also be informative
for trade policy, since the issue of liberalisation of public procurement has been and
continues to be on the agenda of the WTO and other international organisations. Policy makers have long recognised that home-biased procurement may a7ect industrial
relocation. Liberalisation of public procurement has been the object of a number of
EC Directives in the context of the implementation of the Single Market, as well as of
the Government Procurement Agreement in the context of the WTO Uruguay Round.
In its oHcial assessment of the e7ects of the Single Market programme, the European
Commission has for instance acknowledged that the liberalisation of public procurement may lead to “the rationalisation of Community production on a smaller number
of sites” (Emerson et al., 1988, p. 53). As more conventional forms of protectionism are being whittled away, biased procurement thus receives increasing attention in
international policy fora.
Our paper is structured as follows. In Section 2, we set out the theoretical model
and derive three testable propositions. These propositions are tested empirically on
input–output data for EU countries in Section 3. Section 4 concludes.
2. Theory
We explore the impact of home-biased public expenditure on international specialisation in two settings that have become benchmark models of the “new trade theory”
and the “new economic geography”. Our theoretical analysis con4rms that Baldwin’s
neutrality proposition holds for the perfectly competitive sectors, but it shows that
home-biased procurement does a7ect specialisation in increasing-returns monopolistically competitive sectors. We 4rst use the “new trade theory” setting to investigate
whether home-biased public expenditure can attract industrial activity to the home country, and then we turn to a “new economic geography” model to explore the impact of
home-biased procurement on agglomeration and dispersion forces.
2.1. Public expenditure in a static model of international specialisation
In this section we extend the model developed in Helpman and Krugman (1985,
Section 10.4) by introducing government demand. This allows us to investigate the
e7ects of home-biased government procurement on international specialisation.
The basic structure of the model is as follows. We assume two homogeneous factors
of production, generically labelled l and k; two countries, indexed by i = 1; 2; and
three sectors: X , Y , and Z. Two sectors are perfectly competitive (Y and Z) and one
is monopolistically competitive (X ). 1 Sector Z will serve as numNeraire. Production
1
For maximum comparability of our 4ndings with those of the relevant “home-market e7ect” literature,
we work with a general equilibrium model, assuming that there is a suHcient number of goods to yield a
factor-price equalisation set of full dimensionality. This is further explained below.
M. Br3ulhart, F. Trionfetti / European Economic Review 48 (2004) 851 – 881
855
technologies di7er across sectors but are identical across countries. Sectors Y and Z are
subject to a linearly homogeneous production function and operate under perfect competition. The average and marginal cost functions associated with these technologies
are cY (w; r) and cZ (w; r), where the arguments are the remuneration to l and k. The X
sector produces a di7erentiated commodity using a technology that requires a 4xed cost
f(w; r) and a constant marginal cost m(w; r). It is assumed that the functions m(w; r)
and f(w; r) use factors in the same relative proportion. Thus, factor proportions in the
X sector depend only on relative factor prices and not on the scale of 4rms. Since all
X 4rms have identical costs and face the same demand, the optimal level of output is
the same for all 4rms and is denoted by x. The average cost function of the X sector
is cX (w; r) = m(w; r) + f(w; r)=x. Demand functions for factors obtain from the cost
functions through Shephard’s lemma. We denote these demand functions as ls (w; r)
and ks (w; r) with S = X; Y; Z. Further, we assume no factor intensity reversals. Finally,
we adopt the standard assumption that goods Y and Z are traded internationally at
zero costs while good X is traded internationally at a cost of the iceberg type. This
means that for one unit of the X good shipped only a fraction ∈ (0; 1] arrives at its
destination. 2 The total number of X varieties produced in the world, denoted by N ,
is endogenously determined, and so is its distribution between countries. The number
of X varieties produced in country i is ni and we have N ≡ n1 + n2 . The world’s
factor endowment is exogenous and denoted by L and K. Countries’ factor endowments are exogenous, and L1 ≡ L − L2 and K1 ≡ K − K2 . The equilibrium equations
are:
pS = cS (w; r);
S = Y; Z;
(1)
pX (1 − 1=) = m(w; r);
(2)
pX = cX (w; r; x);
(3)
lY (w; r)Yi + lX (w; r)xni + lZ (w; r)Zi = Li ;
kY (w; r)Yi + kX (w; r)xni + kZ (w; r)Zi = Ki ;
i = 1; 2;
i = 1; 2:
(4a)
(4b)
Eqs. (1) and (2) express the usual condition that marginal revenue equal marginal cost
in all sectors and countries. Eq. (3) states the zero pro4t condition in sector X in all
countries. Eqs. (4a) and (4b) state the market clearing conditions for factors in all
countries. These eight equations describe the supply side of the model.
2 Free trade in the perfectly competitive sectors is an assumption of convenience which has the additional
attraction here of assuring comparability with related studies. However, we have shown elsewhere that the
impact of home-biased expenditure on international specialisation (“pull e7ect”) obtains also if we assume
positive trade costs in all sectors (Br%ulhart and Trionfetti, 2002).
856
M. Br3ulhart, F. Trionfetti / European Economic Review 48 (2004) 851 – 881
To close the model, we need to describe the demand side in its two components,
private and public. Households in both countries are assumed to have homothetic
preferences. Speci4cally, we assume Dixit–Stiglitz preferences (i.e., a nested Cobb–
Douglas–CES
utility function) with Cobb–Douglas expenditure shares si (S = X; Y; Z)
=
1, and with an elasticity of substitution of the CES sub-utility equal
and
S Si
to the constant ∈ (1; ∞). Households are taxed in a lump-sum fashion. Homothetic
preferences assure that the distribution of taxation among households does not a7ect
aggregate demand. Maximisation of utility subject to the budget constraint yields households’ demand functions. Aggregating across households gives demand functions for
the di7erentiated good. Country i’s private demand for each variety produced in i is
pX− Pi1− Xi Iid and for each variety produced in j is pX− Pi1− Xi Iid . The price index
Pi =(ni pX1− +nj −1 pX1− )1=(1−) is the price index applicable to country i, Iid =(1−i )Ii
is households’ disposable income, i is a taxation parameter, and Ii is the inner product
between the vector of factor endowments and the vector of factor prices (households
have claims on k). Since pro4ts are zero, Ii is national income. For future reference,
P
≡ Xi (1 − i )Ii .
we de4ne private expenditure on the X good in i as EXi
Governments purchase goods that they use for their subsistence. The balanced budget
requirement assures that expenditure equals tax collection. Tax revenue amounts to
i Ii and is allocated among goods according to the parameter Si (S = X; Y; Z) with
G
3
S Si = 1. Government i’s expenditure on X goods is then EXi ≡ Xi i Ii .
Following Baldwin (1970, 1984) and Miyagiwa (1991), we introduce an exogenously determined parameter that represents governments’ bias in favour of domestically produced goods: "i ∈ [0; 1]. Speci4cally, a proportion "i of government i’s
purchases is reserved to domestic producers. The remainder of government expenditure is allocated eHciently among suppliers from both countries. A large "i therefore means a strong home bias. This simple assumption can represent two widely
used discriminatory practices: (1) the outright exclusion of foreign bidders from domestic public tenders and (2) a domestic-content requirement imposed on foreign
4rms. 4 For clarity of exposition we shall say that government i’s procurement is
“fully liberalised” if "i = 0, “home-biased” if "i ∈ (0; 1], and “wholly home-biased”
if "i = 1. Note that our assumption that home bias appears only in public expenditure and not in private expenditure is one of pure convenience. In fact, all our
results would carry through if we allowed both sources of expenditure to exhibit home
bias, as long as the home bias of public-sector purchasers exceeds that of private
agents. 5
3 Following Baldwin (1974, 1980) and Miyagiwa (1991), we leave the government’s expenditure share
and the taxation parameter exogenous. The same expenditure shares would result from the assumption that
governments produce a public good according to a Cobb–Douglas–CES production function with parameter
shares s and with a constant elasticity of substitution of the CES aggregate equal to ∈ (1; ∞). A constant
per capita tax would instead result from Lindahl-type taxation if we assumed that the government produces
a public good that enters the utility function in a separable way.
4 On the practice of this discriminatory behaviour see Hoekman and Mavroidis (1997).
5 The relevant analytical results are available from the authors.
M. Br3ulhart, F. Trionfetti / European Economic Review 48 (2004) 851 – 881
857
Equilibrium in the product market requires the following equations to hold:
P
P
G
G
pZ (Z1 + Z2 ) = EZ1
+ EZ1
+ EZ2
;
+ EZ2
(5)
pX x = pX1− P1−1 [EXP 1 + (1 − "1 )EXG1 ] + #pX1− P2−1 [EXP 2 + (1 − "2 )EXG2 ]
+ ("1 =n1 )EXG1 ;
(6)
pX x = #pX1− P1−1 [EXP 1 + (1 − "1 )EXG1 ] + pX1− P2−1 [EXP 2 + (1 − "2 )EXG2 ]
+ ("2 =n2 )EXG2 ;
(7)
where # ≡ −1 . Eqs. (5)– (7) close the model. Eq. (5) equates supply and demand
for Z, where demand (r.h.s.) is represented in its four components: country 1’s private
and public expenditure and country 2’s private and public expenditure. Equilibrium
in the X sector requires two equations. Eqs. (6) and (7) represent the equality of
demand and supply for any one variety produced in country 1 and 2, respectively. By
Walras’ law the equilibrium condition for Y is redundant. System (1) – (7) is composed
of 11 independent equations and 12 unknowns (pX ; pY ; pZ ; x; n1 ; n2 ; Y1 ; Y2 ; Z1 ; Z2 ; w; r).
Taking pZ as the numNeraire, the system is perfectly determined.
While all endogenous variables are determined simultaneously, it is useful to inspect
the subsystem (4)– (7) for an intuitive understanding of what shapes the pattern of
international specialisation. Given prices and 4rm scale x, Eqs. (6) and (7) determine
n1 and n2 as functions of private and government expenditure. Then, given n1 and
n2 , the four equations (4a) and (4b) determine the four unknowns Y1 ; Y2 ; Z1 ; Z2 as
functions of factor endowments. This means that, while private and government demand determine international specialisation in the monopolistically competitive sector,
factor endowments determine international specialisation in the perfectly competitive
sectors. Moreover, we can con4rm Baldwin’s neutrality proposition by inspection of
Eq. (5), which shows that world private plus government demand determine world
output of Z (and Y ) but not its international distribution. International specialisation in
these sectors is fully determined by factor endowments via (4a) and (4b). Home bias
in government procurement is therefore inconsequential for international specialisation
in the perfectly competitive sectors. This is Baldwin’s neutrality proposition. His result, originally derived in a small-country partial-equilibrium model, extends also to a
two-country general-equilibrium setting.
A 4nal note on the dimensionality of the model is in order. We have chosen to work
with a general-equilibrium model throughout, both for strict comparability of our own
4ndings and to show that Baldwin’s neutrality proposition extends to such a setup.
Trade costs segment the market for the di7erentiated good. In order to restore full
dimensionality of the factor-price equalisation set we therefore need one more good
than factors, which is why we work with a three-by-two model. Working within the
factor-price equalisation set enhances the tractability of the model, and it makes our
4ndings comparable with those of related papers. Our result in Eq. (8) below, for
instance, is directly comparable to the results on the home-market e7ect literature.
858
M. Br3ulhart, F. Trionfetti / European Economic Review 48 (2004) 851 – 881
We now explore the e7ect of private and government demand on international specialisation. By inspection of system (1) – (7) it is immediate that n1 = n2 is an equilibrium when countries are identical, i.e., when EXP 1 = EXP 2 ; EXG1 = EXG2 , and "1 = "2 . The
nonlinearity of the model prevents us from deriving a simple reduced form. However,
we can 4nd the relationship we are interested in by di7erentiating system (1) – (7) with
respect to changes in private and public expenditure at the equilibrium point n1 = n2 .
It will be convenient to use the following de4nitions:
$i ≡ ni =N;
&P ≡
&iP ≡
EXW ≡ EXP 1 + EXP 2 + EXG1 + EXP 2 ;
EXP 1 + EXP 2
;
EXW
&G ≡
P
EXi
;
+ EXP 2
&iG ≡
EXP 1
EXG1 + EXG2
;
EXW
EXG1
G
EXi
:
+ EXG2
We shock expenditure in such a way that world private and world public expenditure
on each commodity remain unchanged, i.e. ESW (S = X; Y; Z), &G , and &P are held
constant. This implies that the relative prices of commodities will remain unchanged,
and the e7ect on specialisation, if any, is due to changes in a country’s share of
world public and private expenditure &iG and &iP . Technically, this is achieved when we
disturb the equilibrium by d Xi = −d Xj , and by dXi = −dXj . Di7erentiation around
the equilibrium point, where n1 = n2 , yields the following expression:
(1 − # 2 )&P
(1 + #)(1 − # + 2#")&G G
P
d$i =
d&
+
d&i :
(8)
i
(1 − #)2 + 4#"&G
(1 − #)2 + 4#"&G
'1
'2
The 4rst term on the r.h.s. is the e7ect of private expenditure. For convenience we
denote the 4rst coeHcient by '1 . This term shows that, ceteris paribus, large private
expenditure on X results in large domestic output of X (remember that 0 ¡ # ¡ 1). The
second term is the e7ect of government expenditure. For convenience we denote the
second coeHcient by '2 . This term shows that, ceteris paribus, large and home-biased
government expenditure on X results in large domestic output of X . We can thus
formulate a 4rst proposition.
Proposition 1. The country with relatively large home-biased public expenditure on
the di?erentiated good X will, ceteris paribus, be relatively specialised in the production of X .
It is interesting to inspect the relative size of '1 and '2 , because this gives us
the relative size of the impact of private and government demand on international
specialisation. The relative size of '1 and '2 depends on the relative size of &G and
&P . However, if we de4ne b1 ≡ '1 =&P and b2 ≡ '2 =&G , inspection of Eq. (8) shows
that b1 ¡ b2 unambiguously as long as " ¿ 0. The presence of the home bias makes
the impact of public procurement larger than the impact of private demand when both
are appropriately weighted by their size. This result may be expressed in the following
proposition.
M. Br3ulhart, F. Trionfetti / European Economic Review 48 (2004) 851 – 881
859
Proposition 2. The size-weighted impact of home-biased public procurement is larger
than the size-weighted impact of private expenditure.
Henceforth, we refer to Propositions 1 and 2 as the “pull e7ect” of home-biased
public procurement. Proposition 2 can also be viewed in terms of the literature on
home-market e7ects (Davis and Weinstein, 2003; Head and Ries, 2001; Head
et al., 2002). Because of the assumed home bias in public expenditure, the pull e7ect
is not a simple additive component of the standard home-market e7ect. Proposition 2
states that the home-market e7ect of government expenditure, per currency unit spent,
is larger than the home-market e7ect of private demand. This di7erence increases in
the degree of the government’s home bias.
To summarise, we have found that home-biased procurement inQuences international
specialisation in some sectors but not in others, and that its size-weighted impact on
the location of increasing-returns sectors is larger than the impact of private expenditure. In the empirical section we estimate Eq. (8) and 4nd that for the sectors where
home-biased procurement inQuences international specialisation the estimated parameters are such that b1 ¡ b2 .
2.2. Public expenditure in a dynamic model of international specialisation
In “new economic geography” models, international specialisation is shaped by dynamic processes which result from the tension between agglomeration and dispersion
forces. At high trade costs dispersion forces prevail and the industrial activity is evenly
distributed across countries (no industrial agglomeration). At low trade costs agglomeration forces take over, and increasing-returns activity concentrates in a subset of countries (strong industrial agglomeration). In this section we use such a model to study
the e7ect of home-biased government procurement on the likelihood and intensity of
industrial agglomeration.
We use a variant of the model in Trionfetti (2001) which, in turn, is an extension
of the model in Krugman and Venables (1995). The demand side of the model is
the same as in Section 2.1 of this paper, but the supply side is slightly di7erent. We
abstract from factor endowments and assume a single factor of production, labour, and
we can thus also restrict the analysis to two sectors. Employment in each sector and
country is denoted by LSi , where S = X; Z; and i = 1; 2. Analogously, wages in each
sector and country are denoted by wSi . By choice of units we set wZi = 1. As in the
previous section, each variety of the di7erentiated good produced by the X sector is
subject to economies of scale represented by a 4xed cost and constant marginal costs.
The di7erence is that the 4xed and marginal costs are both in terms of a composite
input V , which in turn is produced with labour and X itself. The input requirement
per x units of output is given by V = , + 'x. Each 4rm produces V according to
V = [l=(1 − -)]1−- (X=-)- , where - ∈ (0; 1) represents the importance of the industry’s output as its own intermediate input. Given this technology, the expression for
1−- total costs is TCi = wMi
Pi (, + 'x). Finally, we should note that, unlike in the previous section, private expenditure on X now includes 4rms’ demand for intermediate
860
M. Br3ulhart, F. Trionfetti / European Economic Review 48 (2004) 851 – 881
P
inputs. 6 The expression for private expenditure then becomes EXi
= Xi (1 − i )Ii +
-ni pXi x. The market-clearing equations are the same as in the static model (Eqs. (5) –
P
(7)), provided that we use the appropriate expressions for EXi
.
Concerning the dynamics of the model, it is assumed that labour moves slowly into
(out of) X as the wage in X exceeds (is smaller than) the wage in Z. 7 This assumption
can be formalised with the two di7erential equations L̇X 1 = 1(wX 1 − wZ1 ) and L̇X 2 =
1(wX 2 −wZ2 ), where 1 is an arbitrary constant. 8 The total number of varieties and world
employment in X is constant because world expenditure on manufactures is assumed
constant over time. Individual countries’ employment in X can, however, change over
time. Note that, since world employment in X is constant, the two di7erential equations
can be nicely compacted into one. De4ning ! ≡ wX 1 − wX 2 , and using the fact that
labour market clearing implies that n1 =(n1 + n2 ) = LX 1 =(LX 1 + LX 2 ), we can rewrite the
di7erential equations as
$̇1 = !($1 ; ; "1 ; "2 ; 1 ; 2 ):
(9)
Eq. (9) highlights the fact that ! depends on the state variable $, on trade costs and on
the public procurement parameters. The system is at rest when it reaches an internal
solution or when the X sector is completely agglomerated in one country. Internal
solutions are characterised by wage equalisation across sectors, i.e., wZi = wXi = 1
(i = 1; 2). Complete agglomeration of X activity in one country is associated with
wage inequality, i.e. wX 1 ¿ 1 if $1 = 1, or wX 2 ¿ 1 if $1 = 0. Whether dispersion
forces or agglomeration forces prevail depends on trade costs and on the parameters
of government procurement. It is to this analysis that we turn now.
The economic mechanisms at work can be described in an intuitive way. For the
sake of simplicity suppose that, starting from equilibrium, the system is perturbed by a
random shock that increases the number of 4rms in 1 and decreases it in 2. This initial
perturbation sets in motion four dynamic forces. Two of these forces reinforce the
initial shock, and are therefore called “agglomeration forces”. The other two counteract
the initial shock and are therefore referred to as “dispersion forces”.
1. The expression for total costs reveals that the reduction of P1 (and the increase
of P2 ) caused by the initial increase in n1 and decrease in n2 reduces total costs,
thereby raising 4rms’ potential pro4tability in 1 (and reducing it in 2), and thus
favouring further entry of 4rms in 1 (and pushing 4rms out of the market in 2).
This mechanism, which is known as the forward linkage, tends to reinforce the
initial disturbance and is, therefore, an agglomeration force.
2. The expression for private expenditure shows that increase in n1 (a decrease in
n2 ) increases the expenditure on manufactures produced in 1 via an increase in the
6
Since pro4ts are zero, 4rms’ aggregate expenditure on X is - times 4rms’ aggregate revenue.
Alternatively, it could be assumed that labour is perfectly mobile across sectors and that 4rms move
to the country that yields highest pro4ts. The dynamics resulting from this alternative assumption would be
identical to those we work with.
8 This simplifying practice, which implies myopic expectations, is generally adopted in the literature.
Forward-looking expectations have been studied by Baldwin (2001) and Ottaviano (1999), who have shown
that, when history matters, the direction of motion of the state variables is the same with rational expectations
as with myopic expectations.
7
M. Br3ulhart, F. Trionfetti / European Economic Review 48 (2004) 851 – 881
861
•
η1
0
ηW
ηC
ηE
1
η1
Fig. 1. Public procurement in an economic geography model.
demand for intermediate inputs. This raises potential pro4tability in 1 (and reduces it
in 2), thereby encouraging new entry if 4rms in 1 (and exit in 2). This mechanism,
known as backward linkage, tends to reinforce the initial disturbance and is therefore
an agglomeration force.
3. An increase in n1 shifts the demand faced by each 4rms from country 1 to the
left, and vice-versa in 2. This reduces the potential pro4tability in 1 and increases
it in 2 thus discouraging further entry of 4rms in 1. This mechanism, known as the
competition e?ect, counteracts the initial disturbance and is therefore a dispersion
force.
4. In the presence of home-biased procurement an additional force comes into play.
Inspection of the r.h.s. of (6) and (7) shows that an increase in n1 (decrease in
n2 ) reduces government expenditure on each variety produced in 1 and increases
expenditure on each variety produced in 2, thus reducing potential pro4tability in 1
and increasing it in 2. This, in turn, discourages further entry of 4rms in 1 (while
it encourages entry in 2), and therefore acts as a dispersion force. We refer to this
force as the “spread” e?ect of home-biased public expenditure. The impact of this
force is weakened as public procurement becomes more liberalised, and it disappears
if procurement is fully liberalised.
The relative strength of agglomeration and dispersion forces determines the stability
of the initial equilibrium. When trade costs are high enough dispersion forces always
prevail, while agglomeration forces may dominate at low trade costs. We can illustrate
the e7ect of home-biased procurement on the stability of the equilibria by use of a
phase diagram (Fig. 1). This considers only the case where trade costs are suHciently
low, i.e. such that, if procurement were fully liberalised, agglomeration would occur.
There are at most three equilibria in the set (0; 1). Let us call the closest one
to 0 the “western” equilibrium, the middle one “central” equilibrium, and the one
furthest away from 0 the “eastern” equilibrium. These three equilibria are represented in
Fig. 1 by $W , $C , $E . For simplicity, we assume countries to be identical in every
respect, so that $C = 1=2. In the previous section we were interested in the response
of the equilibrium to asymmetric government demand shocks. Here we are concerned
with the number and stability of the equilibria. 9
9 The results illustrated in Fig. 1, including the threshold values that de4ne “large” or “small” government
procurement and “high” and “low” trade costs, are derived analytically in Trionfetti (2001).
862
M. Br3ulhart, F. Trionfetti / European Economic Review 48 (2004) 851 – 881
Three possibilities emerge.
1. If public procurement is fully liberalised, the central equilibrium is unstable, and
the X sector completely agglomerates in country 1 or 2. This case is depicted by
the solid line.
G
2. If public procurement is home-biased but "i EXi
is small in both countries, the central
equilibrium is unstable but there are two other equilibria with incomplete agglomeration ($W and $E ), which are stable. Therefore, some but not all of the X sector
eventually agglomerates in country 1 or 2. This case is depicted by the dashed line.
G
Furthermore, the distance between $W and $E decreases as "i EXi
increases in both
countries.
G
3. If public procurement is home-biased and "i EXi
is large in both countries, then the
central equilibrium is stable. Therefore, no agglomeration will take place regardless
of trade costs. This case is depicted by the dotted line.
The message emerging from these three cases is that home-biased procurement reduces the likelihood and intensity of industrial agglomeration. It does so in two ways.
First, it may stabilise the central equilibrium, as is clear from a comparison of case 3
with case 1. Second, even if the central equilibrium becomes unstable, the intensity of
agglomeration will relate negatively to the size of home-biased procurement. This is
shown in case 2.
Proposition 3. Large and home-biased public expenditure in one or both countries
reduces the likelihood and intensity of industrial agglomeration. We refer to this
proposition as the “spread” e?ect.
A 4nal clari4cation on the pull e7ect and spread e7ect is in order. Given an initial
equilibrium ($W ; $C , or $E ) an increase of government 1’s share of world expenditure
“pulls” the equilibrium to the right. Therefore, idiosyncratic home-biased government
demand, all else equal, induces specialisation (“pull e7ect”). On the other hand, a
symmetric increase in public expenditure relative to private expenditure stabilises the
equilibrium. A rise in world government expenditure therefore “spreads” industries that
might otherwise be subject to agglomeration forces across countries (“spread e7ect”).
In sum, $i is subject to two forces: the spread e7ect and the pull e7ect. Accordingly,
in a cross-section setting, we would expect the variance of $i to correlate positively
with the variance of &i = &iP + &iG and negatively with &G :
Var($i ) = c1 Var(&i ) + c2 &G ;
(10)
with c1 ¿ 0 and c2 ¡ 0. We can use Eq. (10) to estimate the “spread” e7ect empirically
while controlling for the “pull” e7ect.
3. Empirical evidence
Our theoretical model focuses on the distinction between 4nal expenditure of private agents and 4nal expenditure of the government, assuming that the latter is more
M. Br3ulhart, F. Trionfetti / European Economic Review 48 (2004) 851 – 881
863
home-biased than the former. Input–output tables provide the best statistical information
for a sector-level empirical quanti4cation of these two expenditure types. Our study is
based on a cross-country set of comparable input–output tables which has been compiled by Eurostat and covers up to 11 EU member countries for the period 1970 –1985
in 4ve-yearly intervals. 10 These input–output tables distinguish 18 industrial sectors.
Before we analyse the relationship between countries’ sectoral specialisation and their
relative public and private expenditures, some discussion of the relative home biases
in public and private spending is warranted. Our key assumption is that public-sector
purchasers are more home-biased than private agents. We do not seek to verify this hypothesis here, since evidence in its support has been produced elsewhere. Mastanduno
(1991) and Hoekman and Mavroidis (1997) have provided compelling case studies.
Trionfetti (2000) has compared import shares between public and private purchasers
in the Eurostat input–output dataset that we use here (con4ned to 1985), and he found
that import propensities were lower for public than for private purchasers in 77 percent of all observations. Similarly, the European Commission (1997) reported that,
in 1987, less than two percent of public purchasing of EU countries was awarded
to non-national suppliers, compared to shares ranging between 25 and 45 percent for
private-sector purchases, and it identi4ed public procurement as one of the principal
remaining obstacles to a fully-Qedged Single Market. A fortiori, discriminatory public
procurement must have been a pervasive phenomenon in EU countries during the 1970
–1985 period, which we cover in our empirical study.
3.1. The pull e?ect of public expenditure
The 4rst two propositions derived from our model stipulate that, other things equal,
relatively large discriminatory government expenditure on the product of an increasingreturns industry will result in relatively large domestic output of that product, and that
this “pull e7ect” is stronger when exerted by public expenditure than when it comes
from private demand.
We de4ne industrial specialisation through the following measure (year subscripts
implied):
OUTsi
s OUTsi
OUTdevsi ≡ − where OUTdev ∈ (−1; 1)
i OUTsi
s
i OUTsi
(11)
and where OUT stands for output, s again represents industries and i stands for countries. In order to test the sensitivity of our results to the underlying de4nition of
production, we compute the measure VAdev, which is based on value added data and
constructed in identical fashion to OUTdev. The 4rst summand in Eq. (9) is the empirical representation of $i in our theoretical model, i.e. a country’s share in world
output of a certain industry. For the empirics we subtract from this the country’s share
in total world output as a scaling factor, so as to avoid spurious regression results
10
Eurostat’s input–output tables for more recent years use a less disaggregated sectoral classi4cation for
manufacturing industries and are therefore not considered in this paper. A detailed description of the data
set is given in the Data Appendix.
864
M. Br3ulhart, F. Trionfetti / European Economic Review 48 (2004) 851 – 881
linking expenditure and production shares solely through di7erences in country sizes.
OUTdev and VAdev are centred symmetrically around zero, which represents the point
where a country’s share in the world production of an industry corresponds exactly to
that country’s share in the world’s total manufacturing production. Table 5 reports the
sectors of strongest and weakest specialisation according to VAdev for each country.
Analogously, we construct a measure of idiosyncratic government demand (year
subscripts implied):
DGOVsi
DGOVsi
s
DGOVdevsi ≡ − s
i DGOVsi
i DGOVsi
where DGOVdev ∈ (−1; 1):
(12)
DGOV stands for government expenditure, which we de4ne as the sum of three expenditure headings in the input–output tables: “general public services” (NACE I810),
“non-market services of education and research” (NACE I850), and “non-market services of health” (NACE I890). In addition, we compute a measure of idiosyncratic
private demand DPRIVdev by applying the same formula to the expenditure category
“4nal consumption of households on the economic territory” (NACE F01); and a measure of total idiosyncratic 4nal demand Ddev, which is the sum of public and private
4nal demand. The 4rst summand in the expression for DGOVdev (DPRIVdev) is the
empirical representation of &iG (&iP ) in our theoretical model, while the second term provides the scaling factor needed to eliminate the possibility of contaminating regression
estimates with pure country-size e7ects.
3.1.1. Regressing specialisation on idiosyncratic demand
We can now relate our measure of international specialisation to idiosyncratic government demand. According to our 4rst proposition, a pull e7ect would manifest itself
through a positive relation between these two variables. As a 4rst exercise we have
produced bivariate plots for our four sample years, based on specialisation in output
(Fig. 2) and in value added (Fig. 3). A positive relationship between the two variables
is apparent, but the correlations look rather weak. 11
Our second step was to regress specialisation on idiosyncratic demand. These results
are reported in Table 1. Due to the scaling of our variables we could force the constant
term to zero in all speci4cations. 12 Although our dependent variable is bounded, we
proceeded with a linear speci4cation, since we do not want to make out-of-sample
predictions and since estimations based on limited dependent variable models produced
substantially equivalent results for the relevant data ranges. In Model I of Table 1,
we have regressed specialisation on Ddev in the pooled data. The signi4cant positive
coeHcients con4rm the 4nding of Davis and Weinstein (2003) that home markets
matter for industrial location. Our result is particularly strong since we considered only
11
We 4nd a correlation coeHcient of 0.20 between DGOVdev and both OUTdev and VAdev, pooled
across sample years (see Table 2). These correlations are statistically signi4cantly di7erent from zero at the
99.99 percent con4dence level.
12 Estimations with a constant term or with a variable intercept (i.e. a panel) never produced signi4cant
intercept coeHcients.
M. Br3ulhart, F. Trionfetti / European Economic Review 48 (2004) 851 – 881
year==70
year==75
year==80
year==85
865
OUTdev
0.29
-0.18
0.29
-0.18
-0.32
0.56
-0.32
0.56
DGOVdev
Fig. 2. The pull e7ect: Public expenditure and industry specialisation in output terms.
year==70
year==75
year==80
year==85
VAdev
.26961
-.22451
.26961
-.22451
-0.32
0.56
-0.32
0.56
DGOVdev
Fig. 3. The pull e7ect: Public expenditure and industry specialisation in value added terms.
866
M. Br3ulhart, F. Trionfetti / European Economic Review 48 (2004) 851 – 881
Table 1
Demand deviations and specialisation in production: Pooled runs
Regressand:
OUTdev
Model
Regressors
OLS/beta coe7.
(t stat)
I
Ddev
II
DPRIVdev
0.60/0.61
(10.41)∗∗∗
0.55/0.53
(11.22)∗∗∗
0.08/0.16
(2.94)∗∗∗
0.59/0.63
(7.03)∗∗∗
0.05/0.12
(1.13)
0.51/0.52
(5.77)∗∗∗
0.05/0.11
(0.83)
0.46/0.45
(4.96)∗∗∗
0.08/0.14
(1.45)
0.64/0.50
(5.04)∗∗∗
0.13/0.26
(2.77)∗∗∗
0.51/0.48
(10.14)∗∗∗
0.09/0.18
(3.28)∗∗∗
0.38/0.02
(0.63)
3.59/0.17
(3.86)∗∗∗
0.50/0.47
(8.36)∗∗∗
0.09/0.17
(4.15)∗∗∗
0.38/0.06
(0.86)
0.53/0.51
(12.14)∗∗∗
0.07/0.15
(3.09)∗∗∗
0.04/0.36
(10.66)
DGOVdev
III
1970: DPRIVdev
DGOVdev
IV
1975: DPRIVdev
DGOVdev
V
1980: DPRIVdev
DGOVdev
VI
1985: DPRIdev
DGOVdev
VII
DPRIVdev
DGOVdev
GOVBIAS
DGOVdev ∗ GOVBIAS
VIIIa
DPRIVdev
DGOVdev
Lagged dependent var.
IX
DPRIdev
DGOVdev
NETEXPORT
VAdev
R2
OLS/beta coe7.
(t stat)
R2
No. obs
0.37
627
0.47/0.42
(7.13)∗∗∗
0.39/0.33
(6.59)∗∗∗
0.09/0.17
(2.75)∗∗∗
0.33/0.33
(2.69)∗∗∗
0.06/0.14
(1.01)
0.34/0.32
(3.27)∗∗∗
0.01/0.01
(0.07)
0.37/0.33
(3.55)∗∗∗
0.09/0.14
(1.08)
0.54/0.36
(3.47)∗∗∗
0.19/0.31
(3.55)∗∗∗
0.40/0.35
(6.88)∗∗∗
0.10/0.19
(2.84)∗∗∗
0.22/0.09
(2.35)∗∗
0.83/0.08
(1.17)
0.38/0.31
(5.26)∗∗∗
0.10/0.17
(3.74)∗∗∗
0.10/0.03
(0.54)
0.37/0.32
(6.44)∗∗∗
0.09/0.16
(2.75)∗∗∗
0.04/0.32
(8.48)∗∗
0.18
627
No. obs
0.32
627
0.40
162
0.30
162
0.25
159
0.35
144
0.32
560
0.30
412
0.44
621
0.15
627
0.12
162
0.10
162
0.15
159
0.26
144
0.18
560
0.16
412
0.25
621
Notes: See text for de4nition of variables and data. t statistics are White-adjusted. ∗∗∗ = ∗∗ = ∗ : statistically
signi4cant at 1=5=10 percent level.
a The lagged dependent variable is instrumented using lagged DPRIVdev and lagged DGOVdev, and
standard errors are adjusted accordingly (see Stewart and Gill, 1998, p. 209f).
M. Br3ulhart, F. Trionfetti / European Economic Review 48 (2004) 851 – 881
867
demand for 4nal products in our de4nition of “home markets” and could therefore
eliminate the possibility of upward bias due to simultaneity between output and demand
in the case where demand includes expenditure on intermediate products as well as on
4nal goods.
In Model II of Table 1 we have estimated Eq. (8) in the pooled dataset by taking
account separately of the private and public components of idiosyncratic expenditure.
Both coeHcients are positive and precisely estimated, con4rming our Proposition 1.
The coeHcients on private demand deviations (0.55 for output, 0.39 for value added)
are substantially larger than those on public demand deviations (0.08 and 0.09, respectively). These estimated parameters correspond to '1 and '2 of Eq. (8). Recall from
Section 2.1 that in interpreting these coeHcients one ought to keep in mind the di7erent
sizes of private and public demand. It is through b1 = '1 =&P and b2 = '2 =&G that the pull
e7ect of a marginal dollar spent by public and private agents can be estimated. Table 5
shows that on average private demand was roughly ten times the size of government
demand. Precisely, the average share of public expenditure in total 4nal expenditure,
&G , pooled across years in our dataset is 0.106. Our estimated b1 is therefore equal
to 0.61 in the output speci4cation and to 0.44 in the value added speci4cation, while
our estimate of b2 is 0.73 and 0.87, respectively. These results suggest the presence
of a pull e7ect of public expenditure according to our Proposition 2: an extra dollar
spent by government has a stronger e7ect on attracting production in the relevant sector
than an extra dollar spent by private agents. However, we cannot attribute statistical
signi4cance to this result, since the 95% con4dence intervals of b1 and b2 overlap.
In a third step, we have estimated the empirical version of Eq. (8) separately for each
sample year (Models III–VI in Table 1). We 4nd evidence of an increasing tendency
in the pull e7ect of government purchases. Over the period of our sample, therefore,
the impact of (discriminatory) public expenditure on the location of manufacturing
activities seems to have grown.
Fourth, we augmented the basic speci4cation with the variable GOVBIAS, which is
a proxy for the degree of bias in public procurement by industry and country (Table 1,
model VII). This variable is taken from Nerb (1987), who, based on a survey of 11,000
European 4rms in the mid-1980s, reported the percentage of 4rms who considered the
opening of public procurement markets to be “very important”. This variable might
be a7ected not only by the degree of bias of public purchasers in di7erent industries
and countries but also by the size of public procurement. However, the correlation
between those two variables turns out to be very small and statistically insigni4cant
in our dataset (see Table 6). We therefore added GOVBIAS, as well as an interaction term with DGOVdev, to the baseline speci4cation. We 4nd the expected positive
coeHcients on the two bias variables, and the estimated coeHcients on idiosyncratic
public and private expenditure are barely a7ected. This result con4rms that the more
biased public authorities are in their purchasing activities, the stronger is the pull e7ect
of their expenditure.
Fifth, we re-estimated the original model including the lagged dependent variable
(Table 1, model VIII). This is to account for the fact that international specialisation
patterns change slowly. Since OLS yields inconsistent coeHcient estimates in such a
set-up, we instrumented the lagged dependent variable using the lagged explanatory
868
M. Br3ulhart, F. Trionfetti / European Economic Review 48 (2004) 851 – 881
variables and corrected the standard errors accordingly. We 4nd that the lagged dependent variable is statistically insigni4cant, and its inclusion does not signi4cantly
a7ect the estimated impact of the regressors of interest. This suggests that the 4ve-year
intervals in our data are suHciently long to cover the adjustment lag of international
production structures to changed demand patterns.
Finally, we addressed the issue of potential endogeneity. Although our theoretical model implies expenditure shares that are una7ected by supply conditions, in reality a causal link could exist that runs from output shares to expenditure shares.
Speci4cally, if a country’s price index for a particular sector is negatively correlated with that country’s specialisation in the particular sector, then relatively large
domestic sectors may attract relatively large domestic expenditure shares. 13 The impact of such a causal link is ambiguous a priori: any observed home-market e7ects
of expenditure on output shares will be strengthened (weakened) if the price elasticity of sectoral expenditure is larger (smaller) than one. This is an empirical issue.
Whilst sectoral price indices are not available for the countries in our sample, we
have computed net export ratios as a proxy for international price di7erentials, taking account of the fact that net exporters tend to have lower domestic prices than
net importers. These ratios are de4ned for each sector-country-year observation as:
NETEXPORT = (exports − imports)=(exports + imports). The sample correlation coeHcients with NETEXPORT are 0.05 for DPRIVdev and 0.03 for DGOVdev, which are
both statistically insigni4cant. This suggests that a low relative domestic price in a particular sector does not systematically increase or decrease domestic expenditure shares
allocated to that sector. 14 In Model IX of Table 1, we have added NETEXPORT to
the baseline regression, and we 4nd that this reduces the coeHcients on the idiosyncratic demand variables only very slightly. The qualitative results remain una7ected:
both private and public expenditure shares relate positively to production shares, and,
when weighted by the relative expenditure sizes, this e7ect remains stronger for public
than for private expenditure.
3.1.2. Adding endowments and input–output linkages
Our theoretical setup in Section 2 is richer than the empirical speci4cation that we
have estimated so far. Speci4cally, the models incorporate two additional determinants
of international specialisation: factor-endowment di7erences across countries combined
with di7erent factor requirements of sectors, and agglomeration forces based on input–
output linkages among 4rms. We therefore extend the original empirical speci4cation
that was based on Eq. (8) by adding various combinations of the following regressors
(year subscripts implied):
1. PRIMARYintersi = (Primary inputs used=Output)s
∗ (Primary inputs produced=Manufacturing output)i ,
13 We are grateful to an anonymous referee for pointing this out. For a discussion of this point and a
similar empirical strategy to ours, see also Davis and Weinstein (2003, p. 15f.).
14 Conversely, as expected, the correlation between production shares and NETEXPORT is high (0.39 for
output and 0.34 for value added) and statistically signi4cant.
M. Br3ulhart, F. Trionfetti / European Economic Review 48 (2004) 851 – 881
869
2. AGRIintersi = (Agricultural inputs used=Output)s
∗ (Agricultural inputs produced=Manufacturing output)i ,
3. ENERGYintersi = (Energy inputs used=Output)s
∗ (Energy inputs produced=Manufacturing output)i ,
4. CAPITALintersi = (Fixed capital consumption=Output)s
∗ (Capital stock per worker)i ,
5. WAGESHAREintersi = (Wages=Output)s ∗ (Wages=GDP)i ,
6. MANINPintersi = (Manufactured inputs used=Output)s
∗ (Manufactured inputs produced=Manufacturing output)i .
The 4rst 4ve regressors are interaction variables capturing the factor abundance of countries and the factor intensities of industries, in the spirit of Heckscher–Ohlin theory.
The sixth regressor is constructed in order to control for input–output linkages among
manufacturing industries, which can give rise to endogenous geographical concentrations (an “industrial base”) as described in Section 2.2. Details on the construction of
these variables are given in Appendix A.
If factor endowments and input–output linkages are important determinants of industrial specialisation among EU countries, then we would expect to 4nd positive
and signi4cant coeHcients on all of the regressors. Our results for the entire data set,
reported in Table 2, are largely consistent with those theoretical priors. We have experimented with varying speci4cations of the estimating equation as well as with estimation
techniques that take account of potential year-speci4c heteroskedasticity. 15 Almost all
of the estimated coeHcients are positive, and many are statistically signi4cant. The
exception is the variable capturing input–output linkages, which seems very sensitive
to the chosen speci4cation and gives rise to signi4cant positive as well as negative
coeHcients. A comparison of the results in Table 2 with those of Table 1 shows that
inclusion of the additional regressors adds very little to the explanatory power of the
model; R-squares are raised only slightly, and the estimated coeHcients on DGOVdev
and DPRIVdev are very stable. It is of course not unexpected that endowment di7erences explain a small share of observed specialisation di7erences across the relatively
homogeneous countries of Western Europe; nor would it appear surprising that we
struggle to pick up robust evidence of geographical industry clusters based on input
linkages, given that such concentrations of manufacturing activity would more likely
appear in region-level data than in a country-level dataset.
The main aim of this exercise, however, is to test the robustness of the estimated
coeHcients on the variables that represent idiosyncrasies in public and private 4nal expenditure. Our estimated relationships turn out to be remarkably una7ected by the inclusion of any combination of additional control variables. The coeHcients on DGOVdev
and on DPRIVdev are always statistically signi4cantly positive. In the third-last row
of Table 2 we report the relevant bs, that is the coeHcients appropriately scaled for
15 The estimation might be more eHcient if one could account for potential correlation of disturbances
across years. This cannot be done for the entire dataset, due to the unbalanced nature of the panel, but
estimation on the subsample of countries for which we have observations for all four sample years produced
substantially equivalent results.
Model :
Regressand:
I
II
III
IV
V
OLSa
OLSa
OLSa
OLS with panel-corrected
standard errorsb
Feasible GLS (panel
heteroskedasticity)c
OUTdev
VAdev
OUTdev
VAdev
OUTdev
VAdev
OUTdev
VAdev
OUTdev
VAdev
0.54
(11:38)∗∗∗
0.08
(2:86)∗∗∗
1.68
(3:88)∗∗∗
0.38
(6:60)∗∗∗
0.09
(2:72)∗∗∗
1.37
(2:43)∗∗
0.54
(11:52)∗∗∗
0.08
(2:86)∗∗∗
0.38
(6:65)∗∗∗
0.09
(2:72)∗∗∗
0.56
(9:26)∗∗∗
0.08
(2:56)∗∗∗
0.41
(6:52)∗∗∗
0.13
(3:76)∗∗∗
0.56
(14:88)∗∗∗
0.08
(4:56)∗∗∗
0.42
(9:67)∗∗∗
0.13
(6:46)∗∗∗
0.55
(15:15)∗∗∗
0.07
(4:17)∗∗∗
0.39
(9:81)∗∗∗
0.11
(6:16)∗∗∗
2.15
(4:21)∗∗∗
2.72
(5:60)∗∗∗
2.15
(3:92)∗∗∗
2.65
(5:60)∗∗∗
2.01
(3:85)∗∗∗
1.23
(1.06)
2.22
(3:73)∗∗∗
0.37
(0.38)
0.58
(0.48)
2.72
(7:57)∗∗∗
1.47
(1:67)∗
2.91
(7:29)∗∗∗
0.12
(0.17)
0.64
(0.73)
0.26
(0.73)
0.60, 0.73
0.33
627
1.50
(2:27)∗∗
0.43, 0.87
0.17
627
−0:04
(−0:11)
0.61, 0.73
0.36
627
1.28
(1:89)∗
0.43, 0.87
0.18
627
1.49
(2:24)∗∗∗
−0:43
(−1:34)
0.63, 0.69
0.37
555
2.39
(2:55)∗∗∗
−0:73
(−1:84)∗
[0:47; 1:13]∗
0.27
555
1.49
(2:86)∗∗∗
−0:43
(−0:99)
0.63, 0.69
n.a.
555
2.39
(4:11)∗∗∗
−0:73
(−1:50)
[0:47; 1:13]∗
n.a.
555
1.56
(3:05)∗∗∗
−0:39
(−0:92)
0.62, 0.63
n.a.
555
2.48
(4:43)∗∗∗
−0:62
(−1:33)
[0:44; 1:00]∗
n.a.
555
Regressors
DPRIVdev
DGOVdev
PRIMARYinter
AGRIinter
ENERGYinter
CAPITALinter
WAGESHAREinter
MANINPinter
b1 versus b2 d
Adjusted R2
No. Obs.
∗∗∗ = ∗∗ = ∗ :
Statistically signi4cant at 1=5=10 percent level. ∗ 95 percent con4dence intervals do not overlap.
t statistics in brackets.
b Years de4ned as panels. Beck and Katz (1995) adjusted z statistics in brackets.
c Years de4ned as panels. Observations are assumed to be heteroskedastic across panels, but uncorrelated across panels and non-autocorrelated within panels. z
statistics in brackets.
d b and b are the estimated coeHcients on DPRIVdev and DGOVdev, respectively, divided by the relevant share of private/public expenditure in total
1
2
expenditure.
a White-adjusted
M. Br3ulhart, F. Trionfetti / European Economic Review 48 (2004) 851 – 881
Estimation method:
870
Table 2
Demand deviations, supply-side locational determinants and specialisation in production: Pooled runs
M. Br3ulhart, F. Trionfetti / European Economic Review 48 (2004) 851 – 881
871
the relative size of private and public expenditure. In line with our 4rst two theoretical propositions for increasing-returns sectors, b2 is always larger than b1 ; although,
for most speci4cations, this di7erence is not statistically signi4cant. We therefore conclude (a) that both private and government demand idiosyncrasies are a signi4cant factor shaping the patterns of industrial specialisation among EU countries, and (b) that
idiosyncrasies in public-sector demand matter more than those in private-sector demand,
i.e. there is evidence of a pull e7ect from public expenditure.
As a complement to the pooled runs, we have also estimated our model separately
for each of the 18 industries in our sample. Table 3 reports these results. All variables
with names ending on “abund” correspond to the country-level endowment abundance
terms, i.e. the second multiplicand in each expression that de4nes the interaction terms
given above. We 4nd largely plausible coeHcients on our control variables (food sectors
are bigger in countries with abundant agricultural inputs, textiles and leather industries
are smaller in countries with a large industrial base, etc.). There are a number of industries with statistically signi4cant positive specialisation e7ects of public expenditure
(metal goods, motor vehicles, other transport equipment, rubber and plastic, instrument engineering). On the other hand, there are industries where endowments have
signi4cant explanatory power and expenditure shares do not (chemicals, meat products, timber and furniture, paper and printing). If we took our static model literally,
we would attribute the former set of industries to the monopolistically competitive category, whilst the latter industries would be of the perfectly competitive type. However,
some caution is warranted in the interpretation of these results. Most strikingly, we
4nd implausibly negative and signi4cant coeHcients on the public expenditure variable for two industries, electrical goods and beverages. These counterintuitive results
serve as a reminder of the large range of unexplained variation in our specialisation
measures, and they are possibly due to correlation of idiosyncrasies in public expenditure with some other unobserved variable that determines specialisation. This caveat
notwithstanding, the industry-by-industry regressions support the earlier 4nding that
demand deviations have signi4cant pull e7ects on sectoral specialisation among EU
countries.
3.2. The spread e?ect of public expenditure
According to our third theoretical proposition, the share of (home-biased) public
expenditure in an increasing-returns sector will relate negatively to the degree of specialisation of that sector across countries. This spread e7ect of government demand
is not immediately evident in our dataset. Fig. 4 plots industry averages of absolute
specialisation measures (|OUTdev|) on the industry share of public expenditure scaled
to domestic absorption. For each of the four sample years, we 4nd a tight cluster of
observations near the origin and a single outlier far to the northeast. Fitting a linear
regression line to these data yields a statistically signi4cant positive slope coeHcient
(Table 4, Model Ia). This result holds virtually unchanged when, following Eq. (10),
we in addition control for the standard deviation of DGOVdev (Table 4, Model Ib).
Our initial result is thus diametrically opposed to our theoretical proposition that government expenditure attenuates specialisation pressures.
872
M. Br3ulhart, F. Trionfetti / European Economic Review 48 (2004) 851 – 881
Table 3
Demand deviations, Supply-side locational determinants and specialisation in production: Industry runs (dependent variable = VAdev, 35 observations)
M. Br3ulhart, F. Trionfetti / European Economic Review 48 (2004) 851 – 881
year==70
873
year==75
0.09
mean absolute specialisation (output)
430
290
290
310
390
330
310
510
370 430
510
330
230
210
190 270
350
410490
450
250
230
390
370
210 270
410
190
450250
350
490
470
170
470
170
0.00
year==80
year==85
430
0.09
430
290
290
390
410
370
510
330
270
310
450210
230
390
410
270
370
310
330
510
210
450
350
230
470
250
170
190
490
350 250
190 470
170
490
0.00
0.00
0.34
0.00
0.34
share of public expenditure in absorption
Fig. 4. The spread e7ect: Public expenditure and industry specialisation in output terms: All industries (NACE
codes).
It is worth taking a closer look at the data. Fig. 4 shows that the single outlier in
all years relates to the “other transport equipment” sector (NACE 290). If we drop
this observation from the data set, we 4nd the expected negative impact of public expenditure on specialisation in output terms (Fig. 5) and in value added terms
(Fig. 6). These negative relationships are con4rmed by 4tting a linear model to the
censored dataset (Table 4, Models IIa,b). Hence, our data set as a whole appears to
reject the spread e7ect, yet the elimination of a single industry overturns this result in
favour of our proposition.
Is it justi4able to drop NACE 290 from the dataset for the purpose of this exercise?
In principle, since it is the sector that exhibits by far the largest share of public-sector
demand (see Fig. 4), NACE 290 might provide our most reliable observation. On
the other hand, there are good reasons to believe that the allocation of production in
many subsectors of this industry are not primarily driven by market forces. Around two
thirds of output values in NACE 290 are accounted for by aircraft production (including
military). Inspection of the raw data reveals that in all sample years the most specialised
countries (in terms of both OUTdev and VAdev) were the UK and France, while Italy
and Germany are consistently situated at the bottom of the specialisation scale. This
pattern bears a remarkable resemblance with the development of defence production
874
M. Br3ulhart, F. Trionfetti / European Economic Review 48 (2004) 851 – 881
Table 4
The spread e7ect: Public expenditure and the intensity of specialisation (intercept coeHcients not reported)
R egressand:
Mean|OUTdev|
OLS/beta coe7.
OLS/beta coe7. R2
(t stat)
No. obs (t stat)
No. obs
Model Notes
Regressors
Ia
DGOV/Absorption 0.06/0.23
(2:74)∗∗∗
DGOV/Absorption 0.06/0.24
(2:98)∗∗∗
Std. dev. of
0.07/0.21
DGOVdev
(2:00)∗∗
DGOV/Absorption −0:17=−0.26
(−1:68)∗
DGOV/Absorption −0:16=−0.24
(−1:51)
Std. dev. of
0.05/0.13
DGOVdev
(1:23)
DGOV/Absorption −0:02=−0.34
(−1:25)
DGOV/Absorption −0:02=−0.28
(−1:26)
DGOV/Absorption −0:02=−0.29
(−1:28)
DGOV/Absorption −0:04=−0.38
(−1:49)
Ib
IIa
IIb
IIIa
IIIb
IIIc
IIId
•
•
•
•
All industries
Year 4xed e7ects
All industries
Year 4xed e7ects
•
•
•
•
NACE 290 dropped
Year 4xed e7ects
NACE 290 dropped
Year 4xed e7ects
•
•
•
•
•
•
•
•
1970
NACE
1975
NACE
1980
NACE
1985
NACE
290 dropped
290 dropped
290 dropped
290 dropped
∗∗∗=∗∗=∗ Statistically
Mean|VAdev|
R2
0.09
72
0.13
72
0.11
68
0.13
68
0.12
17
0.08
17
0.08
17
0.14
17
0.03/0.11
(1.09)
0.03/0.11
(1.31)
0.12/0.29
(2:16)∗∗
−0:23=−0.29
(−2:05)∗∗∗
−0:19=−0.26
(−1:94)∗
0.08/0.22
(1:69)∗
−0:01=−0.22
(−0:85)
−0:03=−0.34
(−1:47)
−0:03=−0.37
(−1:63)
−0:05=−0.45
(−1:79)∗
0.10
72
0.18
72
0.19
68
0.23
68
0.05
17
0.12
17
0.14
17
0.20
17
signi4cant at 1=5=10 percent level.
across EU countries that resulted from the peace agreements signed after the second
world war. A case can thus be made for considering the high absolute specialisation
index for NACE 290 as the outcome of determinants that are outside our modelling
framework. Based on this argument, we proceeded to work with a censored dataset,
excluding NACE 290. 16
We estimated a linear relationship between the share of government expenditure in
domestic absorption and the intensity of specialisation, allowing for year-speci4c 4xed
e7ects (Table 4, Models IIa,b). The negative and signi4cant coeHcient estimates provide support for the spread e7ect. We also 4nd the expected positive partial correlation
between the variance of government expenditure and the index of specialisation. Finally, we ran the regression separately for each sample year and obtained consistently
negative parameter estimates (Table 4, Models IIIa–d). The year-by-year coeHcients
16
Four-digit data, taken from the OECD’s Comtap database, show that in 1985 production of aircraft
accounted for 67 (66) percent of output in the “other transport equipment” industry in the UK (France),
while the corresponding shares for Germany and Italy were considerably lower at 44 percent and 25 percent,
respectively. This suggests that specialisation patterns in NACE 290 are shaped by the location of aircraft
production, which is highly politicised at the supra-national level.
M. Br3ulhart, F. Trionfetti / European Economic Review 48 (2004) 851 – 881
year==70
875
year==75
0.09
mean absolute specialisation (output)
430
310
390
310
330
430
370
510
230
510
330
210
350 190
410
450
250
490
390
230
270
370410210
450 190250
350
270
490
470
470
170
170
0.00
year==80
year==85
430
0.09
430
390
390
410
370
230
330 510
270
310
210
450
350
490
190
250
510
410
270
370
330 310
210
450
350
230
470
250
470170
190
170
490
0.00
0.00
0.11
0.00
0.11
share of public expenditure in absorption
Fig. 5. The spread e7ect: Public expenditure and industry specialisation in output terms: NACE 290 dropped
(NACE codes).
are statistically signi4cant only in one of eight cases, but, given the small sample size
at the year level and the stability of the estimated coeHcients, our 4nding seems quite
robust. 17
Arguably our test of the spread e7ect could be biased due to unobserved determinants
of the intensity of industrial specialisation that are statistically correlated with public
expenditure shares. This is a valid concern, but we face two problems in trying to
address it. First, theory does not serve as a useful guide to the speci4cation of control
variables in such an exercise. While trade and geography models are useful in predicting
where certain types of industries will concentrate, they do not provide us with priors
on what features make certain industries more or less concentrated across countries.
Second, the conventional empirical method to address such uncertainty in a context
like ours is to introduce dummy variables for unknown panel-speci4c e7ects; but we
17
Controlling for the variance in government expenditure shares across industries does not signi4cantly
a7ect the estimated coeHcients and standard errors on DGOV/Absorption in the year-by-year regression
runs. To test the sensitivity of the result to the scaling of the regressors, we also estimated the model
on the share of public expenditure in total 4nal expenditure (instead of absorption), and found the results
substantially unchanged. All these results are available from the authors.
876
M. Br3ulhart, F. Trionfetti / European Economic Review 48 (2004) 851 – 881
year==70
year==75
mean absolute specialisation (value added)
0.09
430
310
390
330
310
430
370
510
230
510
330
210
350 190
410
450
250
490
390
230
270
370410210
450 190250
350
490
470
0.00
270
470
170
170
year==80
year==85
0.09
430
430
390
390
230
410
370
330 510
270
310
210
450
350
490
190
250
510
410
270
370
330 310
210
450
350
230
470
250
470170
190
170
490
0.00
0.00
0.11
0.00
0.11
share of public expenditure in final expenditure
Fig. 6. The spread e7ect: Public expenditure and industry specialisation in value added terms: NACE 290
dropped (NACE codes).
are constrained in such an exercise by a lack of degrees of freedom. 18 Our empirical
verdict in favour of the spread e7ect must therefore remain a quali4ed one.
4. Conclusions
This paper sheds light on the consequences of home-biased public expenditure on the
pattern of international specialisation and on industrial agglomeration. The theoretical
analysis is based on models in the mould of the new theories of international trade
and economic geography, and the empirical estimations draw on an input–output panel
data set for EU countries.
In the theoretical part, our paper contributes to the literature by studying the location
e7ects of public procurement theoretically in models with both perfectly competitive
and monopolistically competitive sectors and in models with cumulative agglomera18
Estimation of the model with industry as well as year 4xed e7ects in the censored dataset produced
consistently negative coeHcients on public expenditure shares, but statistical signi4cance was never found.
These results are available from the authors on request.
M. Br3ulhart, F. Trionfetti / European Economic Review 48 (2004) 851 – 881
877
tion features. While our analysis con4rms that home-biased procurement is neutral in
perfectly competitive sectors, it suggests that there are “pull” and “spread” e7ects in
monopolistically competitive sectors. The “pull” e7ect means that a country with relatively large home-biased government expenditure on a certain good will tend to host
a relatively large share of world production of that good. The “spread” e7ect means
that symmetrically large home-biased public procurement reduces the likelihood and
intensity of agglomeration of increasing-returns industries.
The existence of pull and spread e7ects is explored empirically in an input–output
dataset for eleven EU countries, covering the period 1970 –1985. To our knowledge,
this is the 4rst empirical investigation of the e7ect of public procurement on international specialisation in a panel data setting with a full industrial cross-section. We 4nd
evidence in support of the pull e7ect. On average, a country with large government
procurement on a good will tend to specialise in the production of that good. We also
4nd that the e7ect of government demand on international specialisation is stronger
than the e7ect of private demand. Further, we 4nd empirical support for the spread
e7ect. Industries that are subject to a relatively large share of public expenditure tend
to be less concentrated across EU countries.
This study is purely positive in nature. Yet, our results may be informative for policy
making, especially in the context of the WTO Government Procurement Agreement and
of the EU’s ongoing e7ort to complete its single public procurement market.
Acknowledgements
The authors gratefully acknowledge 4nancial support from the Swiss National Science Foundation, from the British Economic and Social Research Council (“Evolving
Macroeconomy” programme, ESRC grant L138251002) and from the French Commissariat GNenNeral du Plan (convention no. 5, 2000). We have bene4ted from useful
suggestions made by seminar participants at the Universities of Paris 1 and Paris 12
and by an anonymous referee.
Appendix A. Data
Our input–output data are taken from the Eurostat “National Accounts ESA” series.
We have data for 18 NACE two-digit manufacturing sectors, up to 11 EU countries
and four sample years (1970, 1975, 1980 and 1985). The country coverage in each
sample year can be gleaned from Table 5. We did not use more recent data, since the
available level of sectoral disaggregation in data after 1985 is signi4cantly lower. The
data for 1970 and 1975 were converted into deutschmarks at current exchange rates,
and those for 1980 and 1985 were converted into current ECUs. The sample contains
up to 630 country-industry-year observations.
Most of the interaction terms used in the model underlying Tables 2 and 3 are based
on variables taken from the input–output database:
• PRIMARYinter: “Primary inputs used” is de4ned as the value of goods from NACE
industries I010–I150 (agricultural and mineral products, power) that are used as
878
M. Br3ulhart, F. Trionfetti / European Economic Review 48 (2004) 851 – 881
Table 5
Descriptive statistics
Country
Year
Share of private
4nal demand in
manufacturing
output
Share of public
4nal demand in
manufacturing
output
Sector of
strongest
specialisationa
Sector of
weakest
specialisationa
Belgium
70
75
80
0.349
0.338
0.344
0.020
0.019
0.021
Instrument
engineering and
other manuf.
Machinery
West Germany
70
75
80
85
0.273
0.274
0.273
0.252
0.040
0.046
0.041
0.041
Motor vehicles
Tobacco products
Denmark
70
75
80
85
0.385
0.356
0.324
0.321
0.041
0.044
0.047
0.040
Meat products
Motor vehicles
Spain
75
80
85
0.425
0.378
0.370
0.013
0.024
0.030
Leather, footwear
Machinery
France
70
75
80
85
0.330
0.334
0.328
0.325
0.029
0.030
0.028
0.040
Other transport
equipment
Tobacco products
Italy
70
75
80
85
0.379
0.344
0.314
0.300
0.013
0.019
0.017
0.024
Leather, footwear
Tobacco products
Netherlands
70
75
80
85
0.354
0.337
0.325
0.271
0.023
0.018
0.025
0.029
Tobacco products
OHce machines
Portugal
80
0.331
0.016
Textiles, clothing
Machinery
U.K.
70
75
80
85
0.347
0.289
0.275
0.284
0.050
0.049
0.066
0.076
Tobacco products
Leather, footwear
Ireland
70
75
85
0.552
0.464
0.363
0.011
0.019
0.019
OHce machines
Motor vehicles
Luxemburg
70
0.616
0.013
Rubber, plastic
OHce machines
Sample
Average
70
75
80
85
0.398
0.351
0.321
0.311
0.027
0.029
0.032
0.037
(n.a)
(n.a)
a Calculated
on the basis of value added data (VAdev) in most recent available sample year.
VAdev
OUTdev
MANINPinter
WAGESHAREinter
CAPITALinter
ENERGYinter
AGRIinter
PRIMARYinter
GOVBIAS
DPRIVdev
DGOVdev
Ddev
Ddev
DGOVdev
DPRIVdev
0.77∗
0.12∗
0.61∗
0.04
0.20∗
−0:01
0.54∗
0.02
0.24∗
0.10
0.02
0.16∗
0.10
0.09
0.35∗
0.20∗
0.42∗
0.17∗
0.12∗
0.01
0.05
0.02
−0:04
0.75∗
0.40∗
−0:04
0.09
0.01
−0:003
−0:02
0:08
0.09
0.17∗
0.08
−0:003
0.01
0.02
*Statistically signi4cant at the 1 percent level.
GOVBIAS
0.05
0.06
0.09
0.03
0.13∗
0.02
−0:05
PRIMARYinter
AGRIinter
ENERGIinter
CAPITALinter
WAGESHAREinter
MANINPinter
0.24∗
0.10
0.05
0.13∗
0.07
0.21∗
0.18∗
0.02
0.01
0.12∗
0.13∗
0.29∗
−0:002
−0:05
0.65∗
0.33∗
0.18∗
0.08
−0:07
0.15∗
0.19∗
M. Br3ulhart, F. Trionfetti / European Economic Review 48 (2004) 851 – 881
Table 6
Correlations
879
880
•
•
•
•
M. Br3ulhart, F. Trionfetti / European Economic Review 48 (2004) 851 – 881
intermediate inputs in the 18 manufacturing sectors in the home country. “Primary
inputs produced” is the total value of output of NACE industries I010–I150 that is
produced in the home country and used as an input in one of the manufacturing
industries.
AGRIinter: “Agricultural inputs used” and “produced” are de4ned equivalently, but
restricted to NACE industry I010 (agriculture, forestry and 4shing).
ENERGYinter: “Energy inputs used” and “produced” are de4ned equivalently, but
restricted to NACE industries I020–I150 (power and mineral products).
WAGESHAREinter: This is based on the heading “gross wages and salaries” in
the input–output tables (NACE F010). Coverage of this variable over countries and
years is incomplete.
MANINPinter: “Manufactured inputs used” is de4ned as the value of goods from the
18 NACE industries I270–I510 (manufacturing) that are used as inputs for production
in those sectors in the home country. “Manufacturing inputs produced” is the total
output of NACE industries I270–I510 that is produced in the home country and used
as an input in one of those industries.
For the construction of CAPITALinter we used data on “4xed capital consumption”
in the input–output table (NACE F080), and values for “capital stock per worker” are
taken from the Penn World Tables. Coverage of this variable over countries and years
is incomplete.
Descriptive statistics and correlations are given in Tables 5 and 6.
References
Baldwin, R.E., 1970. Nontari7 Distortions of International Trade. Brookings Institution, Washington, DC.
Baldwin, R.E., 1984. Trade policies in developed countries. In: Jones, R., Kenen, P. (Eds.), Handbook of
International Economics. North-Holland, Amsterdam.
Baldwin, R.E., 2001. The core–periphery model with forward-looking expectations. Regional Science and
Urban Economics 31, 21–49.
Beck, N., Katz, J.N., 1995. What to do (and not to do) with time-series cross-section data. American Political
Science Review 89, 634–647.
Branco, F., 1994. Favoring domestic 4rms in procurement contracts. Journal of International Economics 37,
65–80.
Br%ulhart, M., Trionfetti, F., 2002. Testing trade theories when expenditure is home biased. Mimeo., University
of Lausanne.
Davis, D.R., Weinstein, D.E., 2003. Market access, economic geography and comparative advantage: An
empirical assessment. Journal of International Economics 59, 1–23.
Emerson, M., Aujean, M., Catinat, M., Goybet, P., Jacquemin, A., 1988. The Economics of 1992: The E.C.
Commission’s Assessment of the Economic E7ects of Completing the Internal Market. Oxford University
Press, Oxford.
European Commission, 1997. Single Market Review: Public Procurement. Kogan Page, London.
Head, K., Ries, J., 2001. Increasing returns versus national product di7erentiation as an explanation for the
pattern of U.S.–Canada Trade. American Economic Review 91, 858–876.
Head, K., Mayer, T., Ries, J., 2002. On the pervasiveness of home market e7ects. Economica 69, 371–390.
Helpman, E., Krugman, P., 1985. Market Structure and Foreign Trade. MIT Press, Cambridge, MA.
Hoekman, B.M., Mavroidis, P.C. (Eds.), 1997. Law and Policy in Public Purchasing: The WTO Agreement
on Government Procurement. University of Michigan Press, Ann Arbor, MI.
M. Br3ulhart, F. Trionfetti / European Economic Review 48 (2004) 851 – 881
881
Krugman, P., Venables, A.J., 1995. Globalization and the inequality of nations. Quarterly Journal of
Economics 110, 857–880.
La7ont, J.-J., Tirole, J., 1991. Auction design and favoritism. International Journal of Industrial Organization
9, 9–42.
Martin, P., Rogers, C.A., 1995. Industrial location and public infrastructure. Journal of International
Economics 39, 335–351.
Mastanduno, M., 1991. Do relative gains matter? America’s response to Japanese industrial policy.
International Security 16, 73–113.
Mattoo, A., 1996. The government procurement agreement: Implication of economic theory. The World
Economy 19, 695–720.
McAfee, R.P., McMillan, J., 1989. Government procurement and international trade. Journal of International
Economics 26, 291–308.
Miyagiwa, K., 1991. Oligopoly and discriminatory government procurement policy. American Economic
Review 81, 1321–1328.
Naegelen, F., Mougeot, M., 1998. Discriminatory public procurement policy and cost reduction incentives.
Journal of Public Economics 67, 349–367.
Nerb, G., 1987. The completion of the internal market: A survey of European industry’s perception of the
likely e7ects. Research on the Costs of Non-Europe: Basic Findings. European Commission, Brussels.
Ottaviano, G.I.P., 1999. Integration, geography and the burden of history. Regional Science and Urban
Economics 29, 245–256.
Stewart, J., Gill, I., 1998. Econometrics, 2nd Edition. Prentice-Hall, Englewood Cli7s, NJ.
Trionfetti, F., 2000. Discriminatory public procurement and international trade. World Economy 23, 57–76.
Trionfetti, F., 2001. Public procurement, economic integration, and income inequality. Review of International
Economics 9, 29–41.
Vagstad, S., 1995. Promoting fair competition in public procurement. Journal of Public Economics 58,
283–307.
Weichenrieder, J.A., 2001. Public procurement in the presence of capital taxation. Regional Science and
Urban Economics 31, 339–353.